Scholarlink

Considering that current antimalware, typically based on signature-based approaches, are not able to identify threats whose signatures are not present in the antiviral database, in this paper we propose a method to identify malware based on deep learning, in particular on convolutional neural networks to identify whether an application is malicious. A distinctive feature of the proposed method is the ability to explain the reasons why the classifier predicts whether an application is malware or trusted, in fact in addition to the binary prediction, the proposed method is able to select the opcodes of the identified application that according to the model are symptomatic of the malicious behavior, thus providing a kind of explainability. Experimental results have shown satisfactory results, thus demonstrating the effectiveness of the proposed method.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

An Explainable Method for Malware Detection Through Convolutional Neural Networks

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Francesco Mercaldo,
  • Paul Tavolato,
  • Antonella Santone,
  • Fabio Martinelli

摘要

Considering that current antimalware, typically based on signature-based approaches, are not able to identify threats whose signatures are not present in the antiviral database, in this paper we propose a method to identify malware based on deep learning, in particular on convolutional neural networks to identify whether an application is malicious. A distinctive feature of the proposed method is the ability to explain the reasons why the classifier predicts whether an application is malware or trusted, in fact in addition to the binary prediction, the proposed method is able to select the opcodes of the identified application that according to the model are symptomatic of the malicious behavior, thus providing a kind of explainability. Experimental results have shown satisfactory results, thus demonstrating the effectiveness of the proposed method.